Constructing Skill Trees for Reinforcement Learning Agents from Demonstration Trajectories

By George Dimitri Konidaris, Scott Kuindersma, Andrew Barto, R. Grupen. 2010

116 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

APRIL: Active Preference-learning based Reinforcement Learning (2012, positive): “The human expert can directly demonstrate some optimal or nearly-optimal behavior, speeding up the acquisition of an appropriate reward function and/or the exploration of the RL search space through inverse reinforcement learning [23], learning by imitation [5], or learning by demonstration [19].”
Challenges in Shared-Environment Human-Robot Collaboration (2013, background): “Various existing algorithms are capable of decomposing tasks into representations agreeable for the internal processing of an autonomous agent [12]–[15].”
Autonomous Skill Acquisition on a Mobile Manipulator (2011, methodology): “In particular, an algorithm called CST (Konidaris et al. 2010) has recently been shown to be capable of acquiring skills from demonstration trajectories on a mobile robot.”
Hierarchical Approaches (2012, methodology): “The method is further developed to build skill trees faster from a set of sample solution trajectories Konidaris et al (2010).”
Programming by Feedback (2014, background): “RL-ILO resumes the celebrated Inverse Reinforcement Learning (IRL) approach (Abbeel, 2008; Konidaris et al., 2010), except for one key difference.”
Learning grounded finite-state representations from unstructured demonstrations (2015, methodology): “For this reason, recent work has aimed at automating the segmentation process [Butterfield et al., 2010; Chiappa and Peters, 2010; Grollman and Jenkins, 2010a; Konidaris et al., 2012, 2010; Kulic et al., 2009].”
Automatic Construction of Temporally Extended Actions for MDPs Using Bisimulation Metrics (2011, methodology): “Other work is focused on learning options from demonstrations [Konidaris et al., 2010].”